A non-invasive approach to estimate the energetic requirements of an increasing seabird population in a perturbed marine ecosystem

There is a growing desire to integrate the food requirements of predators living in marine ecosystems impacted by humans into sustainable fisheries management. We used non-invasive video-recording, photography and focal observations to build time-energy budget models and to directly estimate the fish mass delivered to chicks by adult greater crested terns Thalasseus bergii breeding in the Benguela ecosystem. Mean modelled adult daily food intake increased from 140.9 g·d−1 of anchovy Engraulis capensis during incubation to 171.7 g·d−1 and 189.2 g·d−1 when provisioning small and large chicks, respectively. Modelled prey intake expected to be returned to chicks was 58.3 g·d−1 (95% credible intervals: 44.9–75.8 g·d−1) over the entire growth period. Based on our observations, chicks were fed 19.9 g·d−1 (17.2–23.0 g·d−1) to 45.1 g·d−1 (34.6–58.7 g·d−1) of anchovy during early and late provisioning, respectively. Greater crested terns have lower energetic requirements at the individual (range: 15–34%) and population level (range: 1–7%) than the other Benguela endemic seabirds that feed on forage fish. These modest requirements – based on a small body size and low flight costs – coupled with foraging plasticity have allowed greater crested terns to cope with changing prey availability, unlike the other seabirds species using the same exploited prey base.


Results
Time activity budget in relation to breeding stage. Over a total of 51 days, 374 greater crested tern nests were video monitored during incubation and 240 nests during early chick provisioning (hereafter "early provisioning"). These videos provided duration estimates for 1,138 incubation foraging trips and 1,747 early provisioning foraging trips. Over a 16-day period of focal observations, 31 chicks that had left the nest cup (hereafter "mobile chicks") were monitored during late chick provisioning (hereafter "late provisioning"), which provided duration estimates for 252 foraging trips.
Foraging trips were longer during incubation than during both the early-or late-provisioning periods (Fig. 1A). Incubating adults spent an average of 4.73 h (95% CI 4.51-4.97) away from their nest per trip and performed 1.52 trips·d −1 (1.46-1.58, Fig. 1A,B). Foraging trips during early provisioning were shorter (1.83 h, 1.76-1.90), allowing more trips (4.08 trips·d −1 , 3.88-4.29) than during incubation (Fig. 1B). As a result, the total time spent away from the nest during incubation and early provisioning was similar (Fig. 1C). During late provisioning, when chicks are generally left alone so both adults can forage at once, the mean number of trips per parent per day (4.57 trips·d −1 , 3.97-5.26) was similar to early provisioning (Fig. 1B). In contrast, the mean duration of each foraging trip was longer (2.24 h, 2.02-2.48), resulting in an increase in the time each parent spent away from the chick (Fig. 1C).
Sensitivity analyses showed that variation in adult body mass and prey calorific value had the largest effect on modelled estimates of DFI during all breeding stages (see Supplementary Information S1 and Table S2).
Estimating chick DFI from photo-sampling, video-recording and focal observations. The mean (95% CI) mass of anchovies brought to the chick during early provisioning was 4.4 g (3.9-4.9, n = 126), which

Discussion
Using a combination of different non-invasive methods, this study presents the first estimates of the time budget and linked energy expenditure of a population of breeding greater crested terns. Our results are in agreement with predictions of central-place foraging models, which indicate that adults should increase the amount of energy delivered to chicks over the chick growth period and so raise their own energy expenditure through increased foraging 13,38 . Small chicks were fed anchovies of a size appropriate to their smaller gape, whereas mobile chicks received anchovies ca 20% heavier. Overall, the amount of fish required daily to feed an adult and chick greater crested tern was 3-7 times lower than for other Benguela endemic species relying on the same prey base ( Table 2). A small body size, combined with a highly efficient flight mode and an aptitude for finding food efficiently contribute to lowering the energy budget of greater crested terns. These factors may help to explain why this species' status remains favourable while populations of other Benguela endemic seabirds relying on the same prey base are decreasing.
The use of non-invasive methods for assessing energy expenditure. Uncertainties in reconstructing time-energy expenditure can derive from several sources, including the inaccuracy of activity durations 39 , the estimated cost for each behaviour, and thermoregulatory costs. For terns in particular, these parameters may lack precision as energetic investigations on these birds have so far been limited to small numbers of individuals of only a few species 40 . For example, the model used to estimate flight costs may misrepresent energy expenditure compared to more empirical estimates [40][41][42] . The use of animal-borne data loggers (e.g. GPS, accelerometers)  could overcome this limitation, providing precise time-budget data on different at-sea behaviours (e.g. continuous flapping, gliding, hovering and diving) and estimates of their associated energy expenditure 43 . However, we favoured non-invasive methods as animal-borne data loggers can affect bird condition and behaviour 16 , and because greater crested terns are highly sensitive to human disturbance 44 . Furthermore, the approach used in this study can provide better population-level inference than data logger studies, which usually rely on small sample sizes 13,45 .
Observed feeding rates in our study were limited to delivered prey. However, prior to feeding their chick, provisioning adults may be forced to perform specific behaviours which require additional energetic expenditure. Terns are often the target of inter-and intra-specific kleptoparasitism as they bring prey to the colony in their bill 46,47 . This can result in loss of prey (up to 3.2 g·d −1 of anchovies for interspecific kleptoparasitism) and/or additional energy costs to counter kleptoparasitic attacks 48 . Accordingly, provisioning adults may have to compensate for the food lost in this way, with implications for their energy expenditure 49 ; however, this interaction is poorly understood and few studies can account for the energy expenditure linked to kleptoparasitism in models.

Implications at the population level of low individual energetic requirements. The recent
decreases in seabird populations in the Benguela ecosystem suggest that updated estimates of food consumption are needed to account for energy partitioning in the management of the purse-seine fisheries, with which predators compete for prey 24,31,50 . Modelling approaches are increasingly being implemented to study seabird-fishery competition 23 , including studies to predict the smallest forage fish biomass needed to sustain seabird productivity over the long term 51 . To provide an overview of seabird energetic needs, it is particularly important to account   for species body size, clutch size, and number of fledging days. These needs can then be extrapolated to a broader ecosystem level by accounting for the total population breeding in the system. A comparison of the energetic demands with the other three Benguela endemic seabirds that rely on forage fish, illustrates that the biomass of forage fish needed by breeding greater crested terns at present is much lower than that needed by the other populations (Table 2). Greater crested tern chicks require ~3 kg of anchovy to fledge, compared to ~17 kg of anchovy for an African penguin chick 52 ~10 kg for a Cape gannet chick 28 and ~6 kg for a Cape cormorant chick (T. Cook unpublished data). With approximately 15,000 pairs breeding in the Benguela ecosystem, the whole population requires ~2,800 kg·d −1 of anchovy, which equates to ~133 times less than the Cape gannet population and ~37 times less than the Cape cormorant population breeding in the region ( Table 2). Breeding African penguins, despite a recent decrease in numbers 33 , require ~13 times more food than greater crested terns (Table 2). Thus, their modest energetic requirements may be a key component allowing greater crested terns to cope in a changing and highly exploited environment.
In animals like seabirds, that must travel large distances to secure prey, costs of transport can constitute a large portion of the daily energy budget. Compared to other species of the guild of Benguela ecosystem seabirds specialised on forage fish, the cost of flight per unit of body mass and time in greater crested terns is low (Table 2). Consequently, the overall cost of flight per individual and per time unit in this species is 4-5 times lower than in the other volant seabirds of this guild ( Table 2). In part, this can be attributed to their wing morphology. Like other tern species, greater crested terns have long (90-115 cm) 53 , narrow, pointed wings with low wing loading. This makes them efficient at the slow, agile flight needed when searching for food 54 . Terns are capable of rapid turning, swooping, hovering, vertical take-off and soaring 40 , all with relatively low energy expenditure. Their capacity to explore the marine environment efficiently may help explain why greater crested terns appear more successful than the Benguela ecosystem's other seabird species at coping with decreased food availability.
In the northern Benguela, the population of sardine has been depleted since the early 1970s, and there has been little if any compensation by anchovy, forcing seabirds there to consume low-quality prey such as bearded goby Sufflogobius bibarbatus 55 . In contrast to the declining African penguin population, the small population of greater crested terns (~1,200 pairs), which also relies on bearded goby in Namiba 54 , has remained stable, suggesting an ability to cope when switching to low-quality prey 56 . Terns in the North Sea were found to be most vulnerable and sensitive to sandeel exploitation, presumably as a consequence of their specialized diet, small foraging range and inability to increase parental foraging effort when prey becomes scarce 25 . In contrast, greater crested terns breeding in the Benguela ecosystem could buffer these limitations due to their flexible diet, which includes ca. 50 different prey species 34 and their low fidelity to breeding sites, which are believed to be chosen depending on the local availability of prey immediately preceding the breeding season, rather than by philopatry 32 . In addition, the recent major decrease of migrant tern populations to the Benguela ecosystem (e.g. common tern Sterna hirundo 57 ) may have led to reduced interspecific competition with surface-gleaning seabirds, providing more resources for this resident tern species. In this context, the greater crested terns' low energy requirements combined with their ability to switch to alternative prey provide a great advantage, highlighting the apparent species-specific responses to shifting foraging conditions, which seem to favour the greater crested tern in this ecosystem.
In conclusion, this study shows that greater crested terns have relatively low energy requirements at both the individual and population level, when compared to other seabirds breeding in the Benguela ecosystem that rely on the same resources. These low energy requirements appear to contribute to their recent increase in this exploited ecosystem. Further studies implementing detailed knowledge of the energetics, prey demands and demography of the Benguela's endemic seabirds are needed to understand the apparent differences in their food requirements and assist the development of conservation planning for the threatened seabird species breeding in the region 58,59 .

Measuring time-budget and feeding rates from video-recording and focal observations.
Foraging trip durations and offspring feeding rates of breeding greater crested terns were assessed on Robben Island (33°48′S, 18°22′E), in South Africa's Western Cape Province, using non-invasive video recordings of nest-cup activities during early provisioning ( Figure S1). All methods were approved by the Department of Environmental Affairs (RES2013/24, RES2014/83, RES2015/65) and the animal ethics committee of the University of Cape Town (2013/V3/TC).
Greater crested tern chicks become mobile and leave the nest cup after approximately four days 53 . Thus, we monitored individual chicks banded with engraved colour rings using binoculars and a hide (distance 10-30 m) to determine foraging trip durations and feeding rates during late provisioning. Observations and recordings were made from February to May during three breeding seasons (2013, 2014 and 2015). See Supplementary Information S1 for details on these observations. Video recordings were analysed using VLC media player (VideoLAN project). Three breeding stages were recognised: incubation (during which time, any prey brought to the colony are only used for courtship), early provisioning (the mean week when chicks are provisioned in the nest cup), and late provisioning (the period when adults provision mobile chicks, which typically gather in crèches). Greater crested terns do not forage at night 60 , but our cameras were not always able to capture useable footage from first light or after sunset. Therefore, if birds on focal nests had already left by the start of filming at dawn, or not returned to the nest by the time our cameras could no longer operate due to low light levels, we used nautical twilight as a proxy of their departure and arrival times 61,62 . Nautical twilight is defined when the centre of the sun is 12° below the earth's horizon 63 . The time of twilight on a given date at each colony was obtained from www.timeanddate.com. Estimating chick DFI from photo-sampling. Prey carried by greater crested terns returning to the breeding colony to feed chicks were recorded as part of a program monitoring tern diet 34 . Prey were photographed using a non-invasive photo-sampling technique, allowing for an accurate determination of fish species and standard length 64 For anchovy, we converted estimated fish lengths to mass using a yearly species-specific regression (see Supporting Information S1 and Table S3).

Time-energy budget models.
Time-energy budget models were built for adult greater crested terns to calculate the amount of food that individuals needed to consume daily to rear their progeny in a season (daily food intake -DFI, g·d −1 ). Specific input values shown in Table 3. Two main behaviours were identified: flying and resting at the colony. Precise time-budget data on at-sea behaviour can be identified using activity recorders such as accelerometers 43 . Due to their small size and sensitivity to disturbance, such data is lacking for almost all tern species. Thus, greater crested terns were assumed to be flying the entire time they were away from the colony. This assumption is supported by the fact that, while foraging, greater crested terns do not rest at the sea surface, diving events are infrequent and dives last only a few seconds at most (pers. obs.). Budgets were based on the bioenergetic model elaborated by Grémillet et al. 6 . By considering the duration (D) and metabolism per time unit (M) of each activity daily energy expenditure (DEE, kJ·d −1 ) for adults was defined as: DEE was then converted into adult DFI. Anchovy make up ~65% of the prey species consumed by greater crested terns in the Western Cape 34 but since one of our aims was to compare observed estimates of chick DFI to our model results, for the purpose of the model we assume that anchovy makes up the entire diet (but see Supplementary Information S1). Using the mean (±SD) calorific value (Cp) of 6.22 ± 0.65 kJ·g −1 (wet mass) [65][66][67][68][69] and an assimilation efficiency 37 (Ea) of 0.77 ± 0.34, we calculated adult DFI (g·d −1 ) as: We took adult DFI to represent the total energetic needs during each incubation period. For each of the early-and late-provisioning phases, we estimated total adult DFI as the sum of the fish needed to sustain their own expenditure (DFI), as derived from their time-activity budget, and the amount needed for chick maintenance and growth. Greater crested tern chicks' energetic requirements have not been measured before. Chick energetic requirements were thus estimated by fitting an allometric regression to published data on 10 larid species 37 ( Figure S2). This regression yielded a distribution for the total amount of energy metabolized until fledging (TME, kJ) in relation to asymptotic chick mass (A = 370 g, Table 3): where α is the distribution for the estimate of the allometric regression intercept (posterior mean = 539.5) and β is the distribution for the estimate of the slope parameter (posterior mean = 37.3). Mean chick daily metabolizable energy intake (MEI) (kJ) over the fledging period (40 days) was thus calculated in relation to days taken to fledge (F): We used a breeding success of 0.59 chicks fledged per pair and a fledging period of 40 days 70 (Table 3) to estimate a daily chick mortality rate (CMR) by assuming that nests fail at random through time:  We then used the resulting survival function ( Figure S3) to estimate total adult DFI (TDFI) for each of the early-provisioning (p = 1) and late-provisioning (p = 2) phases as: Metabolic rates of different activities undertaken by the adults were taken from the literature ( Table 3). We used a basal metabolic rate (BMR) of 6.73 W kg −1 derived from respirometry 71 , 2 × BMR as an estimate of the cost of resting at the colony 72 and estimated the cost of flying in greater crested terns (as 5.2 × BMR) with the software Flight 1.25 73 using a wingspan of 1 m 53 , a wing aspect ratio of 10.4 (from the sooty tern Sterna fuscata) 73 and a body mass of 390 g 53 . This software uses aerodynamic modelling, species-specific body mass and dimension to calculate the energetic cost of flying. Terns may use alternative flight modes to continuous flapping (vertical take-off after a dive, hovering over the water in search for prey or gliding) and incur different flight costs depending on the flight mode or the wind field (wind speed and direction). However, we assumed that greater crested terns were flying continuously during their time away from the colony, that the time spent using alternative flight modes was marginal and that overall, greater crested terns experienced an equivalent proportion of different wind speeds and directions. Flight cost (35.6 W·kg −1 ) was thus calculated as the average between the minimum (31.8 W·kg −1 ) and maximum (39.5 W·kg −1 ) power to fly using continuous flapping. Food requirements for the other Benguela endemic seabirds were collected from previous studies ( Table 2).  (Table 3) where data were normally distributed, we used normal priors with observed means and SDs. Where data were expected to be positive-only with positively-skewed errors (e.g. duration data) we used gamma priors with the observed means for the shape parameter and rate = 1. For the allometric regression between TME and asymptotic chick mass, we used uninformative priors 76 with N(0, 10 −7 ) for means (where 10 −7 is precision) and U(1,500, 4,500) for the residual standard error (σ), with the precision specified as σ −2 .
To calculate chick DFI estimated from fish mass recorded by photo-sampling, we used the MCMC method described above to fit a gamma regression with a log-link function to estimate the mean (±95% CI) mass of anchovy returned to the colony by breeding stage (early provisioning = 1, late provisioning = 2) from n = 755 photographs. The mean (±95% CI) number of prey delivered to offspring by breeding stage from n = 274 events recorded on video or during focal observations, the mean (±95% CI) foraging trip duration, and the mean (±95% CI) number of offspring feeds per day (feeding rate) by breeding stage (incubation = 1, early provisioning = 2, late provisioning = 3) were also estimated using gamma regressions with a log-link functions. For the gamma regressions, we used uninformative priors, N(0, 10 −7 ) for the estimated coefficients in the linear predictor and U(0, 100) for the shape parameter. The observed chick DFI was calculated by multiplying the posterior distributions for anchovy mass and number of prey delivered.
For all parameters, we modelled means ±95% Bayesian credible intervals (CI) using three MCMC chains (150,000 samples, burn-in of 50,000 and no thinning). All models unambiguously converged (all R values < 1.01). See Supporting Information S2 for model code.